基于BLEU指标的机器翻译译文差异对比——以China Daily新闻为例

A BLEU-Based Comparative Study of Machine Translation Output Variations—A Case Study of China Daily

摘要: 随着全球化进程的加速，机器翻译技术在跨语言交流中扮演着至关重要的角色。本文以China Daily新闻为源文本，选用Google Translate、DeepL、有道翻译、微软翻译四款机器翻译工具以及新兴翻译工具ChatGPT对中文和英文新闻进行翻译，并采用BLEU指标对译文进行量化分析。研究重点探讨不同翻译工具在处理中文和英文新闻时的表现差异，研究这些工具在处理不同源语言新闻时是否存在显著差异。结果表明，尽管不同翻译工具在BLEU值上的差异不显著，ChatGPT在中译英任务中表现略优于其他机器翻译工具，但在英译中任务中并未展现明显的优势。本文通过量化分析不同翻译引擎在新闻翻译中的表现并为实际应用中翻译工具的选择提供了实证依据。

Abstract: With the acceleration of globalization, machine translation technology plays a crucial role in cross-linguistic communication. This study takes China Daily news articles as the source texts and selects four mainstream machine translation tools—Google Translate, DeepL, Youdao, and Microsoft Translator—along with the emerging tool ChatGPT to translate Chinese and English news. The BLEU metric is applied to conduct a quantitative analysis of the outputs. The research focuses on the performance differences of these tools when handling Chinese and English news, examining whether significant variations exist depending on the source language. Results show that although the BLEU score differences among the tools are not substantial, ChatGPT performs slightly better than other machine translation systems in Chinese-to-English tasks, while it does not demonstrate a clear advantage in English-to-Chinese tasks. By quantitatively analyzing the performance of different translation engines in news translation, this paper provides empirical evidence and offers practical guidance for the selection of translation tools in real-world applications.